CVE-2026-34159 Vulnerability Analysis & Exploit Details

CVE-2026-34159 Vulnerability Summary

CVE-2026-34159: llama.cpp is an inference of several LLM models in C/C++. Prior to version b8492, the RPC backend's deserialize_tensor() skips all bounds validation when a tensor's buffer field is 0. An unauthenticated attacker can read and write arbitrary process memory via crafted GRAPH_COMPUTE messages. Combined with pointer leaks from ALLOC_BUFFER/BUFFER_GET_BASE, this gives full ASLR bypass and remote code execution. No authentication required, just TCP access to the RPC server port. This issue has been patched in version b8492.

CVE-2026-34159 References

External References

• https://github.com/ggml-org/llama.cpp/commit/39bf0d3c6a95803e0f41aaba069ffbee26721042
• https://github.com/ggml-org/llama.cpp/pull/20908
• https://github.com/ggml-org/llama.cpp/security/advisories/GHSA-j8rj-fmpv-wcxw

CWE Common Weakness Enumeration

CAPEC Common Attack Pattern Enumeration and Classification

• Buffer Overflow via Environment Variables CAPEC-10 This attack pattern involves causing a buffer overflow through manipulation of environment variables. Once the adversary finds that they can modify an environment variable, they may try to overflow associated buffers. This attack leverages implicit trust often placed in environment variables.
• Overflow Buffers CAPEC-100 Buffer Overflow attacks target improper or missing bounds checking on buffer operations, typically triggered by input injected by an adversary. As a consequence, an adversary is able to write past the boundaries of allocated buffer regions in memory, causing a program crash or potentially redirection of execution as per the adversaries' choice.
• Buffer Manipulation CAPEC-123 An adversary manipulates an application's interaction with a buffer in an attempt to read or modify data they shouldn't have access to. Buffer attacks are distinguished in that it is the buffer space itself that is the target of the attack rather than any code responsible for interpreting the content of the buffer. In virtually all buffer attacks the content that is placed in the buffer is immaterial. Instead, most buffer attacks involve retrieving or providing more input than can be stored in the allocated buffer, resulting in the reading or overwriting of other unintended program memory.
• Client-side Injection-induced Buffer Overflow CAPEC-14 This type of attack exploits a buffer overflow vulnerability in targeted client software through injection of malicious content from a custom-built hostile service. This hostile service is created to deliver the correct content to the client software. For example, if the client-side application is a browser, the service will host a webpage that the browser loads.
• Filter Failure through Buffer Overflow CAPEC-24 In this attack, the idea is to cause an active filter to fail by causing an oversized transaction. An attacker may try to feed overly long input strings to the program in an attempt to overwhelm the filter (by causing a buffer overflow) and hoping that the filter does not fail securely (i.e. the user input is let into the system unfiltered).
• MIME Conversion CAPEC-42 An attacker exploits a weakness in the MIME conversion routine to cause a buffer overflow and gain control over the mail server machine. The MIME system is designed to allow various different information formats to be interpreted and sent via e-mail. Attack points exist when data are converted to MIME compatible format and back.
• Overflow Binary Resource File CAPEC-44 An attack of this type exploits a buffer overflow vulnerability in the handling of binary resources. Binary resources may include music files like MP3, image files like JPEG files, and any other binary file. These attacks may pass unnoticed to the client machine through normal usage of files, such as a browser loading a seemingly innocent JPEG file. This can allow the adversary access to the execution stack and execute arbitrary code in the target process.
• Buffer Overflow via Symbolic Links CAPEC-45 This type of attack leverages the use of symbolic links to cause buffer overflows. An adversary can try to create or manipulate a symbolic link file such that its contents result in out of bounds data. When the target software processes the symbolic link file, it could potentially overflow internal buffers with insufficient bounds checking.
• Overflow Variables and Tags CAPEC-46 This type of attack leverages the use of tags or variables from a formatted configuration data to cause buffer overflow. The adversary crafts a malicious HTML page or configuration file that includes oversized strings, thus causing an overflow.
• Buffer Overflow via Parameter Expansion CAPEC-47 In this attack, the target software is given input that the adversary knows will be modified and expanded in size during processing. This attack relies on the target software failing to anticipate that the expanded data may exceed some internal limit, thereby creating a buffer overflow.
• Buffer Overflow in an API Call CAPEC-8 This attack targets libraries or shared code modules which are vulnerable to buffer overflow attacks. An adversary who has knowledge of known vulnerable libraries or shared code can easily target software that makes use of these libraries. All clients that make use of the code library thus become vulnerable by association. This has a very broad effect on security across a system, usually affecting more than one software process.
• Buffer Overflow in Local Command-Line Utilities CAPEC-9 This attack targets command-line utilities available in a number of shells. An adversary can leverage a vulnerability found in a command-line utility to escalate privilege to root.

Other 5 Recently Published CVEs Vulnerabilities

• CVE-2026-5311 – A security flaw has been discovered in D-Link DNS-120, DNR-202L, DNS-315L, DNS-320, DNS-320L, DNS-320LW, DNS-321, DNR-322L, DNS-323, DNS-325, DNS-3...
• CVE-2026-34872 – An issue was discovered in Mbed TLS 3.5.x and 3.6.x through 3.6.5 and TF-PSA-Crypto 1.0. There is a lack of contributory behavior in FFDH due to im...
• CVE-2026-34750 – Payload is a free and open source headless content management system. Prior to version 3.78.0 in @payloadcms/storage-azure, @payloadcms/storage-gcs...
• CVE-2026-34749 – Payload is a free and open source headless content management system. Prior to version 3.79.1, a Cross-Site Request Forgery (CSRF) vulnerability ex...
• CVE-2026-34748 – Payload is a free and open source headless content management system. Prior to version 3.78.0 in @payloadcms/next, a stored Cross-Site Scripting (X...